Joseph T. Fennell1, S. M. S. M. Rajib1, Muhammad Usman2, Ahtisham Akbar2, Hamza Khan2, Saad Khan2, Tim Beale4, Julien Lamontagne-Godwin3, Pablo González-Moreno3,5, Abdul Rehman2 and Rene P. Breton1
1Jodrell Bank Centre for Astrophysics, School of Physics and Astronomy, The University of Manchester, Manchester M13 9PL, UK
2CABI Data Gunj Baksh Road, Satellite Town, PO Box 8, Rawalpindi, Pakistan
3CABI, Bakeham Lane, Egham, Surrey, TW20 9TY, UK
4CABI, Nosworthy Way, Wallingford, Oxfordshire, OX10 8DE, UK
5ERSAF. University of Cordoba, 14014, Spain
The use of remote sensing techniques for large-scale monitoring of invasive plant species offers huge promise for better understanding their ecology. A new generation of optical and radar satellites with global coverage and freely available data, combined with increasing computational resource and faster machine learning methods, provide new and exciting opportunities for monitoring and mapping at medium to low spatial resolution. At the same time, consumer and prosumer Unmanned Aerial Vehicles (UAVs) have become more readily available and can carry diverse sensor payloads. Using the spread of Parthenium hysterophorous in Pakistan as an exemplar scenario, we present an end-to-end approach for linking ground surveys with remote sensing datasets for monitoring invasive species. First, we designed an extensive ground-survey campaign, capturing 1400 sites across 3 provinces (KPK, Punjab and Sindh) and 22 districts, to provide a representative ‘ground truth’ dataset of Parthenium and similar non-Parthenium habitats. We then made high resolution temporal, spectral and spatial measurements using handheld devices and a novel UAV-mounted multispectral imaging sensor to better understand the time-evolution of the Parthenium spectral signature. Finally, we trained a probabilistic classifier using optical satellite time-series data, allowing both the quantification of Parthenium extent as well as the uncertainty in the prediction. We present the outputs of our exemplar study and discuss the implications for methodological design of future studies and the use of open source tools developed throughout the project.